Animal Learning & Behavior1985, 13 (4), 451-462

Maze patrolling by rats withand without food reward

R. E. FITZGERALD, R. ISLER, E. ROSENBERG, R. OETIINGER, and K. BATIIGDepartment of Behavioral Biology, Swiss Federal Institute of Technology, Zurich, Switzerland

The effects of foodreward on rats' behavior in radial and Dashiell tunnel mazes were examinedin two experiments. In the first, with animals at ad-lib body weights, food reward reduced speedof movement at the food locations, but did not affect the patterns of movement in either maze.Exploratory efficiency in the Dashiell maze was unaffected by foodreward, and spontaneous patroll­ing of the radial maze by the nonrewarded animals was comparable to the behavior, reportedby others, of rats running for food reward on elevated eight-arm mazes. In the second experi­ment, with subjects maintained at 80% of ad-lib body weights, there was some evidence for "win­stay" learning: food-rewarded rats in the Dashiell maze were relatively more active near thefood locations than were the nonrewarded animals, and more rewarded than nonrewarded ratsrevisited all foodlocations in the radial maze. Nonetheless, exploratory efficiency in the Dashiellmaze was unaffected by foodreward, as was patrolling efficiency in the radial maze, which wasagain comparable to that of rats on elevated mazes. The similarity in behavior of rewarded andnonrewarded animals in these mazes implies that the major determinant of their behavior, whetheror not food reward is provided, is a spontaneous tendency to avoid places recently visited.

Early studies with Y- and T-mazes (Berlyne, 1960;Dember & Fowler, 1958; Fowler, 1965) observed thatthe tendency of the rat to alternate between goal arms inmazes is independent of whether the rat is satiated or hun­gry or whether the choices are food rewarded or not. Morerecently, radial arm mazes have become popular as a toolin behavioral research. In such mazes, the rat is trainedto collect food placed at the arm ends; its task is to avoidreentering arms it has previously visited. Rats become ex­tremely proficient at this task-in an eight-arm radialmaze, they typically make an average of more than sevenchoices before reentering an arm (Olton & Samuelson,1976). We have previously described spontaneous non­rewarded patrolling by rats in complex asymmetrical tun­nel mazes (Battig, 1983; Battig & Schlatter, 1979; Nil& Battig, 1981). The first purpose of the present studywas to describe spontaneous radial-maze patrolling, andto compare it to the behavior of rats running for food.

The theoretical question we address in this study is whyrats running for food reward in radial mazes perform soproficiently. This was initially attributed to their prefer­ence for "win-shift" behavior, which is interpreted asa species-specific foraging strategy (Olton, Handelmann,& Walker, 1981; Olton & Schlosberg, 1978). More re­cent studies, however, have emphasized the fact that therat has a strong tendency to follow a shift strategy,whether or not it is food rewarded for doing so (Gaffan& Davies, 1981, 1982; Gaffan, Hansel, & Smith, 1983;Haig, Rawlins, Olton, Mead, & Taylor, 1983). Gaffanand Davies (1981, 1982) could find no evidence for en­hanced arm alternation produced by food reward (win-

R. E. FitzGerald's mailing address is: Ciba-Geigy Ltd.. R-I066.506.CH-4002 Basel. Switzerland.

shift) in radial and T-mazes, and concluded that the mazebehavior of rats is adequately accounted for by a combi­nation of the principle of spontaneous alternation and theprinciple of associative memory, whereby the rat prefersplaces associated with food to those associated with nofood (win-stay; Gaffan & Davies, 1981). Unfortunately,it is difficult in T- and radial mazes to dissociate patrol­ling for food (win-shift) from spontaneous patrolling, sincethe same path must be taken both to collect all food andto completely explore the maze (see Gaffan & Davies,1981, p. 296). In a Dashiell-type maze, by contrast, thereare multiple routes from any given start to any given goallocation; the rat can visit food-rewarded areas of the mazewithout ambulating through areas never associated withfood. If rats do follow a win-shift strategy, we would ex­pect food reward to increase activity in the nonrewardedareas of the Dashiell maze. If they follow a win-staystrategy, however, food reward should reduce activity innonrewarded areas of the maze. The second goal of thisstudy was thus to seek evidence, in radial and Dashiellmazes, for win-shift and/or win-stay behavior.

In Experiment I, we tested rats at ad-lib body weight,in order to measure normal spontaneous behavior. Theincentive value of food was increased in Experiment 2by food depriving the subjects to 80% of ad-lib bodyweight.

EXPERIMENT 1

MethodSubjecl~. Thirtyexperimentally naive3-month-old Wistar-derived

Roman high-avoidance (RHA/Verh) male rats from the Institute'sbreedingcolonywere used. Their weightsrangedfrom 200to 300 gat the start of the experiment. The rats were group housed (10 percage) in Macroloncages in a temperature-and humidity-controlled

451 Copyright 1986 Psychonomic Society, Inc.

452 FITZGERALD, ISLER, ROSENBERG, OETTINGER, AND BATTIG

Figure 1. (a) External view of the maze apparatus; (b) Groundplan of the maze, showing the activity sensor positions (circled) andthe food locations (arrows); (c) Dashiell configuration and radialmaze configuration (produced by inserting barriers into the Dashiellmaze). Note that the food locations in the Dashiell maze correspondto the ends of the arms in the radial maze.

arm radially symmetrical maze by the insertion of barriers, as shownin Figure lb and lc. Within each arm was a choice point, leadingon the left to a short blind alley, and on the right to the angled maincenter through which the rat was placed into the maze. The ceilingand walls formed a single unit, hinged to the wall of the room, whichwas raised from the floor for removal of the animal at the end ofthe session.

Procedure. The rats were divided randomly into two groups of15 animals each. Subjects were given one daily 4.5-min sessionon 12 consecutive days. Because we were concerned about possi­ble carryover effects from the radial to the Dashiell configuration,the Dashiell maze test was carried out first. On Days 1-6, theanimals were tested in the Dashiell configuration, and on Days 7-12,in the six-arm radial configuration. For the food-rewarded group,one 45-mg Noyes pellet was placed at each of six locations in theouter corners ofthe maze (arrows in Figure lb), corresponding tothe arm ends in the radial configuration. The nonrewarded groupnever received food in the maze. At the beginning of a session,the rat was placed into the maze through the door in the maze center.After 4.5 min, the animal was removed by raising the maze, andany feces and urine were removed from the floor with a damp cloth.The order of testing within groups remained the same throughoutthe experiment, but the order of testing the two groups was alter­nated each day.

Behavioral measures and statistics. The variables used todescribe behavior were as follows:

Dashiell: (I) Time to criterion: time in seconds to reach all sixfood locations, or, for animals that failed to reach all locations,total trial duration. (The term "food location" is also used for thenonrewarded group, for which there was no food at these locations.)(2) Explored area to criterion: number of different photocells in­terrupted to reach all six food locations, or for those animals thatfailed to reach all food locations, area covered during the entiretrial (theoretical minimum 21, maximum 42, see Figure I).(3) Activity to criterion: total number of photocell interruptions toreach all six food locations, or to the end of the trial. To excludeactivity counts due to tail flicks, a given photocell had to be inter­rupted for at least 75 msec to be counted. (4) Activity distributionto criterion: percentage of activity counts occurring in the outerhexagonal runway and radial alleys of the maze until all six foodlocations were reached, or until trial end, whichever came first.

Radial: (I) Time to criterion: time in seconds to reach the mid­points of all six arms (defined as the cells in the outer radial run­ways). This criterion, rather than "all food locations visited," wasused in all analyses because a number of animals, particularly non­rewarded animals in Experiment 2, entered all arms but turnedaround and left before reaching the ends (see Table 2). (2) Numberof blind alley entries to criterion: Just after the entry point of eacharm was a left/right choice, leading to a short blind alley or to themain goal arm. The number of entries into the six blind alleys weremeasured until all arms had been visited, or until trial end.(3) Choice stereotypy to criterion: percent frequency of the mostfrequent turn category in the choices until all arms were visited,or until trial end. This was calculated as (x-c)/(I-c)*loo, wherex = relative frequency of the most frequent turn category (- 2,-1, 0, 1,2,3) and c = chance probability = l/number of possi­ble turn directions = 1/6. This formula was taken from Olton andSamuelson (1976), who used it to adjust the probability of a cor­rect choice for chance. Its value can range from 0 (equal distribu­tion of all turn directions between two anticlockwise and three clock­wise) to 100 (all turns the same). (4) Number of repetitions tocriterion: the number of repetitive arm visits occurring until all armswere visited, or total number of repetitionsduring the trial if criterionwas not reached.

Except where noted otherwise, statistical analysis was by repeatedmeasures ANOVA (BMDP program 2V; Dixon, 1983). Where ap­propriate, individual comparisons were made by t test. A sig­nificance level of p < .05 was adopted in all analyses.

RadialDashiell

(c)

(a)

room (23°C, 50% relative humidity). They were on a 12-hlight/dark cycle and were tested during the dark phase. Food andwater were available ad lib until the day prior to experimentation.On this day, and for all subsequent experimental days, food be­came available for only 4h/day, beginning after the last rat in a givencage had been tested. Water was continuously available. On thisfeeding schedule the rats gained weight slowly during the ex­periment.

Apparatus. An external view of the maze and its ground planare shown in Figure I. The enclosed alleys were 8 ern wide and15 em high. Movement of the animal through the maze was mea­sured by infrared beams 4 em above the floor; the 42 locations moni­tored are shown in Figure lb. Sensor outputs were numbered andtimed in a buffer and transferred to a PDP 11/34 minicomputer,which reconstructed the animal's movements in the maze. Forstatistical analysis, the data were shipped to a mainframe computer

. installation. The Dashiell configuration could be converted to a six-

MAZE PATROLLING AND FOOD REWARD 453

Table 1Number of Rats in Experiment 1 Reaching Criterion (All Food Locations) on Each Test Day

Dashiell Radial

Group Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 1 Day 2 Day 3 Day 4 Day 5 Day 6

Food-rewarded 15 15 13 15 15 14 12 11 15 13 14 14(13) (13) (15) (14) (15) (15)

Nonrewarded 15 15 15 15 15 13 12 14 13 15 14 13(14) (15) (14) (15) (15) (15)

Note-Numbers in brackets refer to the number of rats reaching criterion (middle of all arms) in the radial maze.

4 5Day

3

onF .F

21

50

60

o42

80

100

26

o100

CP

E 100I-

m 34~«

Figure 2. Behavior of food-rewarded (filled circles) and non­rewarded (open circles) groups to criterion (all food locationsvisited)in the Dashiell maze in Experiment 1 (animals at ad-lib bodyweights). From top to bottom: time in seconds to criterion; exploredarea, in terms of the number of different sensors activated; activity(number of sensor activations); and activity distribution (measuredas percent activity in the outer part of the maze) in each of the sixconsecutive daily tests. (All values expressed as means ± SEM.)

ResultsIn the Dashiell maze, most, but not all, animals visited

the six food locations during each test session. The num­bers reaching criterion on each test day are given in Ta­ble 1. In the radial maze, several animals that failed tovisit all food locations did enter all arms, but turnedaround and left before reaching the ends. In this experi­ment, the number of animals entering all arms was notsignificantly greater than the number reaching all foodlocations on any test day (by McNemar test; Siegel, 1956)(See Table 1).

Dashiell maze. Figure 2 shows mean values for time,area explored, activity, and activity distribution tocriterion on each day of the experiment. There were sig­nificant session effects for all four variables[Fs(5,140) > 2.9], but no main effect offood reward andno significant reward X session interaction. Both groupswere significantly more active in the outer part of the mazeon Day 1 than on Day 6, by related t test. The only othersignificant difference between Day 1 and Day 6 was anincrease in area explored to criterion in the nonrewardedgroup. This indicates that the food-rewarded rats did notlearn to go straight from one food location to the next.Theoretically, a score of 21 different sensor activationswould suffice to reach all the food locations (i.e., by run­ning from the center of the maze to and around the outerhexagonal runway; see Figure 1). In fact, rather thandecreasing, mean area explored to criterion in the re­warded group varied from about 26 on the first day toabout 33 on the sixth test day. The proportion of activityin the outer part of the maze decreased significantly inboth groups (related t tests), from about 70% on Day 1to about 60% on Day 6. This phenomenon has previouslybeen described as a "loss of centrifugal tendency" (Bat­tig & Schlatter, 1979; Nil & Battig, 1981), and in thepresent experiment was not affected by the presence offood reward.

We also measured overall efficiency of movement inthe maze, independent of behavior to criterion, by mea­suring the number of sensor activations to visit 6, 12, 18,24, 30, 36, and 42 different sensors for the first time (seeNil & Battig, 1981). Only 1 animal from each group in­terrupted all 42 sensors in the first test session, so datafor this criterion and day are not shown; sample size forthis criterion in all subsequent sessions was at least 8 pergroup. As shown in Figure 3, these scores suggest im­provement in exploratory efficiency over the six test ses­sions for both groups. This was confirmed by repeatedmeasures ANOVA on activity count to interrupt 36 differ-

454 FITZGERALD, ISLER, ROSENBERG, OETTINGER, AND BATTIG

onF .F

0

,~ 10~

iJ' 515~~ 0

~100

i 50Q),~

6 0

en 38

J 210 I I I

1 2 4 5Day

Figure 4. Behavior of food-rewarded (filled cirlces) and non­rewarded (open circles) groups to criterion (all arms visited) in theradial maze in Experiment 1 (animals at ad-lib body weights). Fromtop to bottom: time in seconds to criterion; number of blind alleyentries; choice stereotypy, measured as percent frequency of the mostfrequently occuring turn direction (see Method, Behavioral Mea­sures section); and number of repetitive arm visits. All values areexpressed as means ± SEM.

in time to criterion; the rewarded group took longer over­all to visit the food locations than the nonrewarded group[group F(l ,28) = 14.6]. This may have been due simplyto the time spent eating the food pellets. We analyzed thetransition times between adjacent sensors in all parts ofthe maze, and found significantly longer times for thefood-rewarded group, but only for the food locations, asin the Dashiell maze. Both rewarded and nonrewardedgroups were significantly faster to criterion on Day 6 thanon Day 1, and made significantly fewer blind alley en­tries. Neither choice stereotypy nor number of repetitions

onFeF

To42differentsensors

...C~

0u 100>....'S...o«

50

Figure 3. Exploratory efficiency of food-rewarded (F) and non­rewarded (nF) groups in the Dashiell maze in Experiment 1 (animalsat ad-lib body weights); decreasing activity count over days to in­terrupt 24 and more different sensors indicates latent learning (seetext for details). Values represent mean scores ± SEM.

150

To36

~ ~To30

- ~To24

~=r-~~T018To12~To6

1 2 3 4 5 6

Day

ent sensors, showing a significant sessions effect[F(5,140) = 19.5], but no main or interaction effects offood reward. These data indicate that latent learning oc­curred (described by Battig & Schlatter, 1979; Nil & Bat­tig, 1981), resulting in the rats' moving less redundantlyover days through the Dashiell maze, and that this processwas not affected by the presence of food.

Total activity in the Dashiell maze (i.e., total numberof sensor activations during each 4.5-min trial) did notdiffer between the two groups over the 6 test days; in bothgroups, mean activity increased on Days 2 and 3, and thenreturned to baseline levels on Days 4-6 [sessionF(5,140)=6.4].

The only significant intergroup difference was obtainedwith an analysis of the transition times between sensorsat the food locations. These were overall significantlylonger in the rewarded than in the nonrewarded group[mean 3.1 vs. 2.7 sec, group F(l,28) =8.6]. There wereno intergroup differences in transition times at the otherloci (i.e., those loci not directly at the food locations).

Radial maze. Figure 4 shows, for each test session,mean scores for (from top to bottom) time, number ofblind alley entries, choice stereotypy, and number ofrepetitive arm visits to criterion. The only significantdifference between the two groups in the radial maze was

MAZE PATROLLING AND FOOD REWARD 455

onFeF

456Serial posn.

of choice

3 4 5Serial posn,oforiginal choice

o +1 +2 +3Choice posn rei.to previous choice

3

2

2

1

-2 -1

.1

.6

.8

·4,...

.6

.8

.4

.3

.4

.2

.2

to

c.

(a)

(c)

(b)

Figure S. Arm choice behavior of the food-rewarded (F) and non­rewarded (oF) groups in the radial maze in Experiment 1 (animalsat ad-lib body weights), pooled over the 6 test days. (a) Probabilityof a nonrepetitive arm choice as a function of the serial position ofthe choice; probability of a correct choice was above chance for allchoices. (b) Probability of a repetitive arm choice as a function ofthe serial position of the original correct choice; probability of repe­tition was highest to arms chosen early in the sequence. (c) Relativefrequencies of turn categories in the choice sequence to criterion (allarms visited); +2 was the most frequent turn category in bothgroups. The formulae for calculation of (a) and (b) were taken fromOlton and Samuelson (1976). All values are expressed as means ±SEM.

changed signficantly over days, and neither was affectedby food reward.

In order to see how closely spontaneous radial mazepatrolling in this six-arm tunnel maze resembed the be­havior reported by others in eight-arm elevated mazes,we analyzed behavior in terms of three frequently reportedmeasures of radial maze performance. The results areshown in Figure 5. Because there were no significantchanges over days for the repetition measure, and in orderto obtain enough observations for an accurate estimationof behavior, the data for each animal were averaged overDays 1-6. Figure 5a shows the probability of a correct(i.e., nonrepetitive) choice as a function of serial choiceposition, that is, arm choices I to 6, corrected for chance.The probabilities were calculated as in Olton and Samuel­son (1976, p. 106); chance performance was zero. Asreported by Olton and Samuelson, mean probability ofa correct choice decreased in successive choices, but re­mained above chance for all choices. Figure 5b shows therelative probability of a repetition as a function of theserial position of the original choice (corrected for op­portunities), for the first five correct choices; this indi­cates a recency effect, whereby the probability of a repe­tition was highest for arms chosen early in the sequence,as found in eight-arm mazes (Olton and Samuelson).Figure 5c shows the relative frequency of all possible turncategories, measured up to criterion (all arms visited).The most frequent turn was two away from the previouschoice.

Total activity (number of sensor activations) per ses­sion in the radial maze showed no significant changeseither between or within groups over the 6 test days.

DiscussionThe distribution of activity of rewarded rats in the

Dashiell and radial mazes in this experiment could notbe differentiated from that of nonrewarded rats, althoughspeed of movement at the food locations was lower in thefood-rewarded group in both mazes. In the Dashiell maze,food-rewarded animals did not learn to go straight fromone food location to the next; area explored to criteriondid not change significantly from the first to the last testday. Efficiency of exploration of the entire maze and theprocess of latent learning, as measured by the activitycount to reach 36 different photocells, were unaffectedby food reward.

In the radial maze, entries into the blind alleys decreasedrapidly within the first few sessions in both groups. (Thiseffect is not due to the animals' previous exposure to theDashiell configuration, since we have consistently foundthe same pattern of avoidance behavior whether the animalhas had previous exposure to other maze configurationsor is maze naive; Isler, Oettinger, FitzGerald, & Battig,in preparation). Food reward had no effects on arm choicebehavior in terms of choice stereotypy or number of repe­titions to criterion. As reported by others in elevated eight­arm mazes, a recency effect was apparent in the choice

456 FITZGERALD, ISLER, ROSENBERG, OETTINGER, AND BATTIG

behavior of all animals. The most preferred turn direc­tion was two arms away from the previous choice; ineight-arm mazes, entering adjacent arms is apparentlymore likely (Yoerg & Kamil, 1982). Notably, patrollingefficiency in the radial maze was stable over days; therewas no change in number of repetitions to criterion withrepeated testing, and spontaneous patrolling was indistin­guishable from food-rewarded patrolling in this respect.

Overall, the results indicate that rats will spontaneouslypatrol relatively complex mazes in a manner that is vir­tually indistinguishable from that of animals running forfood reward. It is perhaps rather surprising that food re­ward had so little effect on the animals' behavior. By us­ing animals at ad-lib body weights, we were concentrat­ing on normal spontaneous behavior in this experiment.As a strong test of the effects of food reward, we testedanimals deprived to 80% of ad-lib body weights in a sec­ond experiment.

EXPERIMENT 2

In this experiment we used basically the same proce­dure as in Experiment 1, but with animals deprived to80% of ad-lib body weights to increase the incentive valueof the food reward.

One aspect of behavior that we noticed in Experiment 1,but which we did not attempt to quantify, was that allanimals continued to be active after visiting the food lo­cations. In Experiment 2, we looked in detail at behavioroccurring after criterion was reached, that is, with no foodnow present for the rewarded group.

MethodThere were several differences in method with respect to Experi­

ment 1. Three-month-old female, instead of male, RHA/Verh ratswere used (no males were available). Ad-lib body weights at thebeginning of the experiment were between 18~ a~d 200 g. A to.talof 16 animals (8 per group) were used. Beginning 1 week pnorto testing, feeding times were adjusted so tha~ all. animals' bodyweights were reduced by 20%, and were so maintained throughoutthe experiment. The apparatus was the same as described for Ex­periment 1, but movement of the animal through the maze was meas­ured by underfloor electromagnetic field-effect sensors. In order

to obtain as much information as possible on behavior occurringafter criterion had been reached, test duration was increased from4.5 to 10 min, and the number of tests in each maze was increasedfrom six to eight. This procedure was successful; most of the animalsin both rewarded and nonrewarded groups reached criterion a se­cond time, that is, revisited all food locations. We could thereforeanalyze the effects of food reward, not only by comparing the be­havior of rewarded and nonrewarded animals up to criterion (first"round" of activity) as in Experiment I, but also by analyzingchanges in behavior in the food-rewarded group from the first tothe second round, that is, with and without food reward. All otherprocedural details were as described in Experiment 1.

Results and DiscussionDashiell maze. With one exception on Day 2, all

animals in both rewarded and nonrewarded groups visitedthe six food locations on every test day, and most animalsin both groups revisited all food locations within the10 min trial (Table 2). Figure 6 shows behavior in termsof mean time, explored area, activity count, and activitydistribution to criterion on each of the 8 days of the ex­periment. Behaviors during the first and second roundsof activity (i.e., all food locations visited, and all foodlocations revisited) are shown in the left and eight panels,respectively, of Figure 6. Those animals that failed tocomplete a second round were assigned the values reachedat the end of the session. The single animal that failedto complete a first round on Day 2 was assigned, as scoresfor Round 2 on this day, means of its Round 2 behaviorover the other 7 days.

During the first round of activity, the rewarded groupwas significantly faster to criterion, explored a smallerarea, had lower activity, and was relatively more activein the outer parts of the maze than the nonrewarded groupover all days [group Fs(I,14) > 12]. Despite the appar­ent increasing difference between the two groups fromthe start to the end of the experiment, seen in Figure 6,significant session and session x food reward effects oc­curred only for the activity distribution measure[F(7,98)=2.6], apparently because the food-rewardedgroup was relatively more active in the outer part of themaze on Days 3-8 than the nonrewarded animals. Theseresults indicate that the food-rewarded animals in this ex-

Table 2Number of Rats in Experiment 2 Reaching Criterion

(All Food Locations) for the First and Second Time on Each Test Day

Food-Rewarded Group

8 8 8 8 8 7 8 8 8 8 8 8 88 8 8(8) (8) (8) (8) (8) (8) (8) (8)

8 8 8 7 4 5 7 8 8 7 7 62 8 8 8 7(8) (8) (6)(6) (7) (8) (8) (8)

Nonrewarded Group

8 8 8 7 8 7 8 7 8 7 78 7 8 8 8(8) (8) (8) (8) (8) (8) (8) (8)

6 7 5 . 0 4 3 4 3 3 2 37 8 6 72 8(6) (8) (6) (6) (8) (6) (6) (8)

Note _ Numbers in brackets are the number of rats reaching criterion (middle of all arms) in the radial maze.

MAZE PATROLLING AND FOOD REWARD 457

200

~100i=

0

42

cu 34Q)..

<I:26

120

onF .F

~...~

0

40 I I I I I I I I I I1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

Day Day1st

,round' 2nd 'round'

Figure 6. Behavior of food-rewarded (filled circles) and nonrewarded (open cir­cles) groups to criterion (all food locations visited) in the Dashiell maze in Experi­ment 2 (animals at 80% of ad-lib body weights). The left panel shows behavior un­til criterion was reached for the first time; the right panel shows behavior untilcriterion was reached a second time, that is, until all food locations were revisited.From top to bottom: time in seconds to criterion; explored area, expressed as thenumber of different sensors activated; activity to criterion, measured by the num­ber of sensor activations; and activity distribution, expressed as the percentage ofactivity occuring in the outer parts of the maze. All values are expressed as means± SEM.

periment did move more directly from one food locationto the next than the nonrewarded animals, until all foodlocations had been visited.

During the second round of activity, until all the foodlocations were revisited (with no food now present foreither group), behaviorof the two groups did not differsignificantly in terms of time, area explored, or activityto criterion. Relative activity in theouterpartof the maze,however, was significantly higher in the rewarded thanin the nonrewarded group, as it had beenduring the first

round. Repeated measures comparison between Rounds Iand 2 showed that there were no significant changes inbehavior in the nonrewarded animals, but that the food­rewarded grouptook longer, exploreda greaterarea, andhad higher activity counts to revisit the (previous) foodlocations than when they were visitingthem for the firsttime. Relative activity in the outer part of the maze didnot change significantly in the rewarded group fromRound 1to Round 2. It is particularly interesting that thebehavior of the rewarded animals was so similar to that

458 FITZGERALD, ISLER, ROSENBERG, OETTINGER, AND BATTIG

of the nonrewarded animals during the second round:Having consumed all the available food, they behavedmuch like animals that had never found food in the maze,except that they were relatively more active in the outerpart of the maze, where they had previously fed.

Efficiency of patrolling the entire maze, independentof behavior to criterion, improved over sessions as in Ex­periment 1, as measured by the activity count to reach36 different sensors [session F(7,98)=4.8]. There were

no main or interaction effects of food reward on this be­havior. Total activity in the Dashiell maze decreased sig­nificantly over days during this experiment[F(5,140)=6.4], but was not affected by food reward.

Radial maze. At least 7 of the 8 animals in each groupvisited all arm ends (food locations) on every test day.During the second round, as shown in Table 2, more re­warded than nonrewarded animals revisited all armends-significantly more, by Fisher exact probability test,

300 onF .F

1 23456781Day

1st 'round'

2345678Day

2nd 'round'

Figure 7. Behavior of food-rewarded (filled circles) and nonrewarded (open cir­cles) groups to criterion (all armsvisited) in the radial maze in Experiment 2 (animalsat 80% of ad-lib body weights). The left panel shows behavior until criterion wasreached for the first time; the right panel shows behavior until criterion was reacheda second time, that is, until all arms were revisited. From top to bottom: time inseconds to criterion: number of blind alley entries; choice stereotypy, expressed aspercent frequency of the most frequent turn category (see Method, Behavioral Mea­sures section); and number of repetitive arm visits. All values are expressed as means± SEM.

MAZE PATROLLING AND FOOD REWARD 459

Figure 8. Arm choke behavior of the food-rewarded (Ii) and non­rewarded (oF) groups in the radial maze in Experiment 2 (animalsat 80% of ad-lib body weights), pooled over the 8 test days. (a) Pr0b­ability of a repetitive arm choice as a function of the serial positionof the choice; probability of a correct choice was above chance forall choices. (b) Probability of a repetitive arm choice as a functionof the serial position of the original correct choice; probability ofrepetition was highest to arms chosen early in thesequence. (c) Rela­tive frequencies of turn categories in the choice sequence to criterion(all arms visited); 11 (adjacent arm) was the most frequent turncategory in the food-rewarded group. The formulae for calculationof (a) and (b) were taken from Olton and Samuelson (1976). Pointsrepresent mean values ± SEM.

onF.F

2 3 4 5 6Serial posn.

of choice

1 2 3 4 5Serial posn.oforiginal choice

-2 -1 0 +1 +2 +3Choice posn. reI.to previous choice

.1

.6

.8

·4=r

.6

.8

.2

.4

.3

.4

.2

to

to

c.

~

~(a) ~

...8c.

.§(b) :::

I!

(c)

(Siegel, 1956) on Days 5 and 7. Nonetheless, as shownin brackets in Table 2, mostof the nonrewarded animalsdid reenter all arms, reaching at least their midpoints.

Figure 7 shows the behavior of rewarded and non­rewarded groups during the first and second rounds ofactivity in termsof meantime, blindalleyentries, choicestereotypy, and numberof repetitions to criterion. Dur­ing the first roundof activity, until all arms were visitedfor the first time,thefood-rewarded ratsweresignificantlyfaster to criterion than the nonrewarded rats over all 8days[F(1,14)=31.9]. [Therewasalsoa group x sessioninteraction effect for this time measure, F(7,98)=2.2,probably because the rewarded group ran increasinglyfasteron Days 3-6, whereasthe nonrewarded groupdidnot.]Theonlyothersignificant effectof food rewardwason choice stereotypy, which wassignificantly higher over­all in the food-rewarded group[F(1,14)=7.4]. As in Ex­periment 1, the numberof blindalleyentriesto criteriondecreased in all animals over days [F(7,98)= 32.9], andwasnot significantly affected by foodreward. In contrastto Experiment 1, the numberof repetitive arm entriestocriterion decreased over days [F(7,98)=8.3]; both re­warded andnonrewarded groups madesignificantly fewerrepetitions to criterion on Day 8 thanon Day 1 (byrelatedt test). From these results it can be concluded that, as inExperiment 1, food reward did not affect behavior interms of blind alleyentries and repetitive arm entries tocriterion. The increase in choice stereotypy in the re­warded group may have arisen because this group wasrunning faster than the nonrewarded group. Olton, Col­lison, and Werz (1977) reported a similar increase inchoice stereotypy with repeated testing in the elevatedradial maze with food reward; stereotypy was reducedby delaying the animal between choices.

Wealsoanalyzed radial maze behavior in terms of com­monly reported measures of choice performance. Theresults are shown in Figure 8. As in Experiment I, themean probability of a correctchoice decreased in succes­sivechoices, but remained abovechance for all choices;a recency effect was also apparent, whereby the proba­bility of a repetition was highest for arms chosen earlyin the sequence. Neither of these measures was affectedby food reward. Turn behavior did appear to differ be­tweenthe two groups: for the nonrewarded animals, themost frequent turn was two away from the previouschoice,whereas in the food-rewarded animals, it wasoneaway from theprevious choice. Thisdifference, however,was not significant by repeated measures (group x turncategory) ANOVA. It is interesting in thiscontext to notethat the most preferred turn direction of food-deprivedrats in eight-arm maze experiments is oftenalso intoad­jacent arms (Yoerg & Kamil, 1982). Both this effectandthe increased choice stereotypy in these rewarded animalsmay be attributable to their tryingto get aroundthe mazeas quickly and with as little motor effort as possible.However, it is also possible that the rats wereattemptingto reduce memory load (i.e., were using adjacent turn-

460 FITZGERALD, ISLER, ROSENBERG, OETTINGER, AND BATTIG

ing as a response algorithm). To test this possibility, itwould be necessary to artificially interrupt the animals'choice behavior in some way.

During the second round of activity, more animals inthe rewarded group than in the nonrewarded grouprevisited all of the arm ends (Table 2). However, mostof the animals in both groups reentered all arms a secondtime (Table 2, numbers in brackets). The right-hand panelof Figure 7 shows behavior during this second round, untilall arms were reentered.

Over the 8 test days, the rewarded animals were sig­nificantly faster to criterion in Round 2 than the non­rewarded animals [F(1,14)=4.6]. The two groups did notdiffer in number of blind alley entries and choicestereotypy, but the rewarded animals made fewer repeti­tions to criterion than the nonrewarded group in this se­cond round [F(1,14)=8.9].

Comparison of behavior between the first and secondrounds showed that both groups were slower to criterionin the second round than in the first [Fs(1,7) > 22]. Therewas no significant overall change in number of blind al­ley entries in either group from Round 1 to Round 2, butthere were significant session x round interaction effectsin both groups [Fs(7,49) > 4], apparently because, forboth groups, blind alley entries decreased over the 8 daysof testing in the first round but remained constant duringthe second round. Over all 8 days, choice stereotypy wassignificantly lower during the second round than duringthe first in both groups [Fs(1, 7) > 35]. In both groups,the mean number of repetitions during Round 2 was lowerthan that during Round 1 on Days 1 and 2, but onDays 3-8, both groups made more repetitions in the se­cond than in the first round. This was reflected in a sig­nificant session x round interaction for both groups[F(7,49) > 2.4], without significant main group effects.

The results for this second round of activity indicatethat the previous experience of food reward did influencesubsequent nonrewarded maze activity. First, more re­warded than nonrewarded animals revisited all the armends. Second, although the rewarded animals were slowerto visit all arms in the second than in the first round, theywere still faster in Round 2 than the nonrewarded animals.Third, nonrewarded animals made more repetitive visitsto criterion in the second round than the rewarded animals.This effect in the nonrewarded animals may not be relia­ble: we have run replications of the experiment in whichnonrewarded animals patrolled as efficiently as rewardedanimals during the second round.

The parallel changes in choice stereotypy and time tocriterion from first to second round in both groups (timeto criterion increased and choice stereotypy decreased)lends support to the idea that choice stereotypy is at leastpartly determined by running speed (see Magni, Krekule,& Bures, 1979; Olton, Collison, & Werz, 1977). Therewas apparently no simple relationship between choicestereotypy and patrolling efficiency, in terms of repeti­tive arm entries, since choice stereotypy was higher over­all in the rewarded than in the nonrewarded group in

Round 1, but the number of repetitions made by the twogroups was not significantly different. This finding is inagreement with the findings of Magni et a1. (1979), Ol­ton et a1. (1977), and Yoerg and Kamil (1982).

Total activity per session, independent of criterion,changed significantly over days [F(7,98) = 11.8], but wasnot affected by food reward; it increased significantly inboth groups from Day 1 to Day 8.

Overall, the results ofthis experiment indicate that hun­gry rats will minimize the time required to visit foodsources. When given the opportunity (i.e., in the Dashiellmaze), they move more directly between food sources,but when the path available for exploration is the sameas that required to reach the food sources, their efficiencyof patrolling is the same as that of spontaneously explor­ing animals.

GENERAL DISCUSSION

The first purpose of this study was to describe the spon­taneous patrolling behavior of rats in the radial tunnelmaze and compare it with that of food-rewarded rats. Theresults indicate that patrolling behavior in terms ofavoidance of already-visited arms was virtually unaffectedby food reward, whether the animals were at ad-lib bodyweights or were food-deprived to 80% of ad-lib weights.In both experiments, furthermore, patrolling behavior wascomparable to that reported by Olton and Samuelson(1976) for rewarded animals in the elevated eight-armmaze. Several aspects of patrolling behavior did, however,differ between the first and second experiments. In Ex­periment 1, there was no change in patrolling efficiencyover days, whereas in Experiment 2 all animals becamemore efficient with repeated testing. Also, over all ses­sions, the animals in Experiment 2 patrolled more effi­ciently than those in Experiment 1. Turn choice behavioralso differed: the most frequent turn direction in Experi­ment 1 was two away from the previous choice for bothrewarded and nonrewarded groups, whereas in Experi­ment 2, the rewarded group tended to enter adjacent armsmore frequently. Further experiments will be necessaryto determine whether these differences between Experi­ments 1 and 2 are attributable to the higher level of depri­vation of the subjects in Experiment 2. Also, we cannotrule out the possibility that sex is an important factor: malerats were used in Experiment 1, females in Experiment 2.Despite the differences in arm choice behavior betweenthe two experiments, it can be concluded from theseresults that nonrewarded rats will spontaneously patrolthe radial maze as efficiently as rats that are rewardedfor doing so.

The second aim of the study was to seek evidence forwin-shift and win-stay behavior produced by food reward.Win-shift effects of food reward should have been observ­able in the Dashiell maze as a reduction in relative ac­tivity at the food locations. In Experiment 1, activity dis­tribution was unaffected by food reward. In Experiment 2,relative activity was actually higher at the food locations

in the rewarded than in the nonrewarded rats over all ses­sions: this behavior is the opposite of that which wouldbe expected if food reward increases the probability ofshifting. Since win-shifteffects may have disappeared withrepeated testing as the animals learned where the foodsources were located, we also looked in detail at behavioron the first test day. However, there was no significantchange (by related t test) in activity distribution fromRound 1 to Round 2 in either group on this day (nor wasthere any Round l-Round 2 change for any other be­havioral measure on Day 1). On the basis of these results,we conclude that there is no evidence for win-shift be­havior by rats in the Dashiell maze; that is, there is noevidence that the probability of moving to less familiaror less frequently visited locations is increased by foodreward. In the radial maze, providing food reward didnot affect the number of repetitions; that is, animals thatdid not find food in the maze shifted as efficiently asanimals that did find food.

A systematicconfound in the design of the present studywas that all animals were tested in the radial maze afterprevious experience in the Dashiell configuration. Thismeans that when tested in the radial maze for the first time,they were already familiar with the maze apparatus, andhad an expectancy of reward or no reward. This expec­tancy might be expected to produce an increase in shiftby the rewarded animals during the first radial maze test,because, in the immediatelypreceding Dashiell maze tests,they had encountered food pellets which were not replacedafter consumption (i.e., a depletion condition, see Haiget aI., 1983). The data from the first test day in the radialmaze are crucial in this respect: analysis by t test revealedno significant differences in any behavioral variable be­tween the rewarded and nonrewarded groups on the firstradial maze test in Experiment 2. In Experiment 1,however, the rewarded animals made significantly morerepetitions to visit all arms on the first radial maze testday than the nonrewarded animals, suggesting that foodreward, rather than increasing shift behavior, actually in­creased stay behavior. We have subsequently run experi­ments comparing food-rewarded and nonrewarded radialmaze behavior in test-naive animals (Isler et al., in prepa­ration); the results suggest that naive animals do not be­have in exactly the same way as animals with previousmaze experience, but the overall conclusions concerningthe effects of food reward on patrolling are the same.

We use the term "win-stay" to refer to the win-stayassociative principle given in Gaffan and Davies (1982):that naive animals tend to expect food (or no food) in cir­cumstances or places in which they have previously ex­perienced it, and subsequently prefer stimuli associatedwith food to those associated with none. We found evi­dence for win-stay behavior in both Dashiell and radialmazes in Experiment 2. In the Dashiell maze, althoughthere were no differences in behavior between rewardedand nonrewarded rats on the first test day, the rewardedrats were subsequently (over all 8 test days) relativelymore active in the vicinity of the food locations than thenonrewarded rats during Round 1, until all the food had

MAZE PATROLLING AND FOOD REWARD 461

been consumed. Even after all the food had been eaten(i.e., during Round 2), the rewarded animals, althoughthey explored as much of the maze as the nonrewardedanimals, were still relatively more active in the vicinityof the food locations. In the radial maze, there was noevidence for win-stay behavior during the first round, untilall food locations had been visited for the first time: thenumber of arm repetitions and blind alley entries did notdiffer significantly between rewarded and nonrewardedgroups. During the second round, however, more re­warded than nonrewarded animals returned to all six foodlocations at the ends ofthe arms (Table 2). Nonetheless,patrolling efficiency of the rewarded animals did not differover all eight sessions between the first (rewarded) andthe second (nonrewarded) round. In order to assess theimmediate effects of finding food reward in the radialmaze, we also examined changes in behavior fromRound 1 to Round 2 on the first test day. Both rewardedand nonrewarded groups made fewer blind alley entriesin the second than in the first round (by related t test).The nonrewarded animals were quicker in the secondround than in the first and made significantly fewer repe­titions, but the rewarded animals' speed and efficiencyof patrolling did not change significantly from the firstto the second round on this first radial maze test. Thiscould be taken as evidence for win-stay behavior by therewarded rats.

Given that the differences between rewarded and non­rewarded animals are explicable in terms of win-stay' as­sociative learning, the overall similarity of their behaviorimplies that the major determinant of patrolling in thesemazes, whether or not food reward is provided, is a spon­taneous tendency to avoid places recently visited (seeGaffan & Davies, 1981; Haig et al., 1983).

Two aspects of the spontaneous maze patrolling be­havior described here deserve particular emphasis. First,it appears to be genuinely spontaneous: all animals, bothrewarded and nonrewarded, patrolled both mazes on firstexposure to them. Rats running for food reward onelevated radial mazes, by contrast, typically require 1 to2 weeks' shaping and training before they will run relia­bly and efficiently (e.g., Olton & Samuelson, 1976; butsee Bruto & Anisman, 1983, for a description of spon­taneous elevated radial maze patrolling by mice). The se­cond point is that activity levels remain high in non­rewarded animals, even after many days of testing in thesame apparatus (up to 30 consecutive days; Isler, Oet­tinger, FitzGerald, & Battig, unpublished observations).Such behavior does not occur in open-field tests, in whichactivity usually declines with repeated exposure, or inelevated mazes, in which the animal will cease to run ifreward is withheld. Perhaps the resemblance of this tun­nel maze apparatus to the rat's natural burrow habitat(Lore & Flannelly, 1978) is important; certainly, the sit­uation appears to be relatively nonstressful to the rat (defe­cation and urination during testing are extremely rare).

The application of optimal foraging theory to rat be­havior in the laboratory is extremely fruitful (see, e.g.,Fantino & Abarca, 1985; Kamil & Sargent, 1981). Un-

462 FITZGERALD, ISLER, ROSENBERG, OETTINGER, AND BATTIG

fortunately, the spontaneous behavior described here isapparently not easily amenable to such analyses. Theproblem is that optimal foraging deals with discrete in­strumental events (i.e., the collecting of food) that canbe measured and manipulated quite explicitly, whereasthe rats spontaneously running around in these tunnelmazes are apparently collecting information of some kind.Furthermore, in the case of the nonrewarded animals inthis study, the information they are collecting is from anunchanging, familiar environment. O'Keefe and Nadel(1978) assigned a primary role to mismatch as a causalfactor underlying exploration, but acknowledged thatfamiliar stimuli are also explored. They described suchexploration as the animal's "mak[ing] a cursory checkto ensure that nothing has changed" (p. 255). Exactly howone goes about quantifying and manipulating this infor­mation is as much of a problem today as it was 25 yearsago, although further experimental progress should bepossible, based on the significant developments in bothlearning and cognitive map theories that have been madesince then (review in Toates, 1983). We are left with thesuspicion that the optimal foraging behavior demonstratedby rats in mazes may be a special case of a more generalphenomenon, optimal information foraging, whose inves­tigation has been rather neglected.

REFERENCES

BATTlG, K. (1983). Spontaneous tunnel maze locomotion in rats. InG. Zbinden, V. Cuomo, G. Racagni, & B. Weiss (Eds.), Applica­tions ofbehavioral pharmacology in toxicology. New York: RavenPress.

BATTIG, K., & SCHLATTER, J. (1979). Effectsof sex and strain on ex­ploratory locomotion and development of nonreinforced mazepatroll­ing. Animal Learning & Behavior, 1, 99-105.

BERLYNE, D. E. (1960). Conflict, arousal and curiosity. New York:McGraw-Hill.

BRUTO, V., & ANISMAN, H. (1983). Alterations of exploratory patternsinduced by uncontrollable shock.Behavioral & NeuralBiology, 1983,31, 302-316.

DEMBER, W. N., & FOWLER, H. (1958). Spontaneous alternationbe­havior. Psychological Bulletin, 55, 412-428.

DIXON, W. J. (Ed.). (1983). BMDPstatistical software /983. Berkeley:University of California Press.

FANTINO, E., & ABARCA, N. (1985). Choice,optimalforaging, and thedelay-reduction hypothesis. Behavioral & Brain Sciences, 8, 315-330.

FOWLER, H. (1965). Curiosity and exploratory behavior. New York;Macmillan.

GAFFAN. E. A., & DAVIES. J. (1981). The role of exploration in win­shift and win-stay performanceon a radial maze. Learning & Moti­vation, 12, 282-289.

GAFFAN, E. A., & DAVIES, J. (1982). Reward, novelty and spontane­ous alternation. Quarterly JournalofExperimental Psychology, 348,31-47.

GAFFAN. E. A., HANSEL, M. C., & SMITH, L. E. (1983). Does rewarddepletion influence spatial memory performance? Learning & Moti­vation, 14, 58-74.

HAIG, K. A., RAWLINS, J. N. P., OLTON, D. S., MEAD, A., & TAY­LOR, B. (1983). Food searching strategies of rats: Variables affect­ing the relative strength of stay and shift strategies. Journalof Ex­perimental Psychology: Animal Behavior Processes, 9, 337-348.

KAMIL, A. C., & SARGENT, T. D. (Eds.). (1981). Foraging behavior:ecological, ethological and psychological approaches. New York:Garland STPM Press.

LORE, R., & FLANNELLY, K. (1978). Habitatselection and burrowcon­struction by wildRattusnorvegicus in a landfill. JournalofCompara­tive & Physiological Psychology, 92, 888-896.

MAGNl, S., KREKULE, 1., & BURES, J. (1979). Radial maze type as adeterminant of the choicebehavior of rats. Journal of NeuroscienceMethods, 1, 343-352.

NIL, R., & BATTIG, K. (1981). Spontaneous mazeambulation and Hebb­Williams learning in Roman High-Avoidance and Roman Low­Avoidance rats. Behavioral & Neural Biology, 33, 465-475.

O'KEEFE, J., & NADEL, L. (1978). The hippocampus as a cognitive map.Oxford: Clarendon Press.

OLTON. D. S., COLLISON. c., & WERZ, M. A. (1977). Spatial memoryand radial arm maze performance of rats. Learning & Motivation,8, 289-314.

OLTON, D. S., HANDELMANN, G. E., & WALKER, J. A. (1981). Spa­tial memoryand food searchingstrategies. In A. C. Kamil & T. D.Sargent(Eds.), Foraging behavior: Ecological, ethological andpsy­chological approaches. New York: Garland STPM Press.

OLTON, D. S., & SAMUELSON, R. J. (1976). Remembrance of placespast: Spatial memory in rats. Journalof Experimental Psychology:Animal Behavior Processes, 2, 97-116.

OLTON, D. S., & ScHLOSBERG, P. (1978). Food searching strategiesin youngrats: Win-shift predominates over win-stay. Journal ofCom­parative & Physiological Psychology, 92, 609-618.

SIEGEL, S. (1956). Nonparametric statistics for the behavioral sciences.New York: McGraw-Hili.

TOATES, F. M. (1983). Explorationas a motivational and learningsys­tem: A cognitive incentive view.InJ. Archer& L. I. A. Birke(Eds.),Exploration in animalsandhumans(pp. 72-116). London:Van Nos­trand Reinhold.

YOERG, S. 1., & KAMIL, A. C. (1982). Response strategiesin the radialarm maze: Running around in circles. AnimalLearning & Behavior,10, 530-534.

(Manuscript received March 4, 1985;revision accepted for publication September 23, 1985.)